Measuring Contagion with a Bayesian, Time-Varying Coefficient Model

Abstract

To measure contagion empirically, we propose using a Bayesian time-varying coefficient model estimated with Markov Chain Monte Carlo methods. The proposed measure works in the joint presence of heteroskedasticity and omitted variables and does not require knowledge of the timing of the crisis. It distinguishes contagion not only from interdependence but also from structural breaks. It can be used to investigate positive as well as negative contagion. The proposed measure appears to work well using both simulated and actual data.

Keywords: Contagion, Gibbs sampling, Heteroskedasticity, Omitted variable bias, Time-varying coefficient models

JEL Classification: C11, C15, F41, F42, G15